Skip to main content

Overview

AI Builder lets you describe what you need in plain language and have an AI agent configure it for you. It works in two modes:
ModeHow it worksBest for
Quick suggestionsA single LLM call generates the configurationRapid first draft, simple APIs
Advanced builderA ReAct agent uses tools in a loop to build, test, and refineComplex APIs, OpenAPI import, iterative refinement
You can switch between modes at any point. The quick mode creates a starting point; the advanced builder lets you iterate.

Connector Builder

A Connector defines how FIM One talks to an external system — its base URL, authentication, and the specific API actions it exposes. The Connector Builder gives an AI agent 9 tools to build and manage this configuration on your behalf.

Tools

ToolWhat it does
Get SettingsRead the current connector config (URL, auth type, auth config)
Update SettingsChange the connector name, base URL, or auth credentials
List ActionsSee all existing API actions with their methods and paths
Create ActionAdd a new API endpoint — HTTP method, path, parameters, body template
Update ActionModify an existing action (description, schema, response extraction)
Delete ActionRemove an action that is no longer needed
Test ActionSend a live HTTP request for any action and inspect the response
Test ConnectionVerify the base URL is reachable and the credentials are accepted
Import OpenAPIBatch-import up to 50 endpoints from a Swagger 2.x or OpenAPI 3.x spec

Typical workflow

The most common pattern: paste an OpenAPI URL and let the builder do the rest. Example prompt:
“Import the OpenAPI spec from https://api.acme.com/openapi.json, then test the GET /orders endpoint with order_id=12345.”
The builder fetches the spec, creates all actions automatically, fires a test request, and reports the result — all without you touching a form.

Agent Builder

An Agent is a named AI persona with a set of instructions, tools, and (optionally) connectors. The Agent Builder gives an AI agent 6 tools to configure another agent from scratch.

Tools

ToolWhat it does
Get SettingsRead the current agent config (instructions, execution mode, tools, model)
Update SettingsChange name, description, system prompt, execution mode, or suggested prompts
List ConnectorsBrowse all available connectors (attached and unattached)
Add ConnectorAttach a connector so the agent can call its actions as tools
Remove ConnectorDetach a connector (the connector itself is not deleted)
Set ModelSwitch the underlying LLM, or tune temperature and max tokens

Typical workflow

Start with a description and let the builder configure the whole agent: Example prompt:
“Create a Finance Copilot. It should answer questions about orders and invoices using the Acme connector. Use ReAct mode and add 3 suggested prompts for common questions.”
The builder reads the current settings, writes a system prompt, attaches the connector, sets the execution mode, and adds suggested prompts — in a single conversation turn.

How it works

Under the hood, both builders share the same infrastructure as regular agents:
Builder modeMechanism
Quick suggestionsA single LLM inference call generates the full configuration as structured JSON
Advanced builderA ReAct agent loop: Reason → call a builder tool → observe the result → decide next step
The advanced builder is a full ReAct agent that happens to have a restricted toolset — only the 9 Connector or 6 Agent builder tools, no web or computation tools. It reads the current state of the target resource, plans what needs to change, calls the appropriate tools, and verifies the result before declaring it done. This means the advanced builder can handle ambiguity: if the OpenAPI import creates 30 actions but only 5 are relevant, you can tell it “keep only the order-related endpoints” and it will delete the rest.